Table 4.

Comparison of computational complexity and performance of the two deep learning (DL) models. For U-Net, both implementations based on 3 × 3 and 5 × 5 filters were considered.

Model Name	Number of Parameters		Size on Disk	Inference Times/Dataset
	Trainable	Non-Trainable		CPU	GPU
E-Net	362,992	8352	5.8 MB	115.08 s	20.32 s
U-Net (3 × 3)	1,946,338	0	23.5 MB	1211.99 s	39.53 s
U-Net (5 × 5)	5,403,874	0	65.0 MB	1411.69 s	46.21 s